对于现阶段的大语言模型，我的看法很复杂。这篇笔记不打算深究这些想法——我也不想展开这个话题。

不过，虽然我在这里做的并非什么前沿研究，但我的内容确实是用心写给人类看的。想到自己花 10 分钟写下的东西，未来或许能帮几个人节省时间，就让我觉得很值得。

我这么做，可不是为了让别人拿去训练模型，再拿去卖订阅赚钱。

我新版的robots.txt已经屏蔽了多个已知大语言模型爬虫的用户代理。我列出的大多数爬虫都已确认，并且都附上了来源。新闻出版方的 robots.txt 也很有参考价值，因为它们同样会屏蔽大语言模型。

如果你也想这么做，欢迎直接拿我的去用！

# I know this can just be _totally_ ignored by crawlers
# But let's hope they behave well :)

# Code: https://github.com/ellie/notes
# Source: https://darkvisitors.com/

# OpenAI, ChatGPT
# https://platform.openai.com/docs/gptbot
User-agent: GPTBot
Disallow: /

# Google AI (Bard, etc)
# https://developers.google.com/search/docs/crawling-indexing/overview-google-crawlers
User-agent: Google-Extended
Disallow: /

# Block common crawl
# I have mixed feelings on this one, but many models are trained on this data
# It is also used to bootstrap new search indices though
# https://commoncrawl.org/ccbot
User-agent: CCBot
Disallow: /

# Facebook
# https://developers.facebook.com/docs/sharing/bot/
User-agent: FacebookBot
Disallow: /

# Cohere.ai
# https://darkvisitors.com/agents/cohere-ai
User-agent: cohere-ai
Disallow: /

# Perplexity
# https://docs.perplexity.ai/docs/perplexitybot
User-agent: PerplexityBot
Disallow: /

# Anthropic
# https://darkvisitors.com/agents/anthropic-ai
User-agent: anthropic-ai
Disallow: /

# ...also anthropic
# https://darkvisitors.com/agents/claudebot
User-agent: ClaudeBot
Disallow: /